type
status
date
summary
slug
tags
category
password
URL
icon
第一节，课程总览
这节课程介绍了关于智能体的一系列课程计划，从基础概念到高级应用。通过构建不同类型的智能体团队，学习者将学会如何利用这些智能体自动化各种任务，并最终构建出能够优化个人简历以提高面试机会的系统。同时，文章还探讨了智能体自动化相较于传统自动化的优势，强调了模糊输入和输出在当前AI应用中的重要性。
课程内容
在这门课程中，我们会讨论很多有趣的概念。我想先强调几项内容，方便你了解接下来会看到什么。我们将讨论角色扮演和智能体，它们能够专注使用工具并相互合作。我们还会深入探讨保护措施的重要性以及记忆如何提升智能体的性能。我们会讨论智能体合作的各种方式，不仅是顺序和层级合作，还包括异步合作。
我们将会构建许多不同的智能体团队，这些团队将由多个具有明确角色的AI智能体组成。别担心，如果你还不明白这些概念，这正是这门课程的目的所在。我们将从概述开始，构建一个简单的研究和写作团队，然后逐步构建更复杂的团队，如客户支持团队、客户推广团队、规划代理系统和财务分析代理系统。最后，我们将构建一个最复杂的团队，一个能够为任何工作岗位量身定制简历的团队，提高你获得面试的几率。因此，请务必坚持到最后，因为接下来会越来越有趣。
我们将以一个全栈工程师职位的招聘信息为例。你会看到其中有很多有趣的内容。仔细阅读职位描述，你会发现他们基本上是在寻找一位全栈开发人员，需要熟悉前端和后端技术，并具备编写API和数据库经验的能力。这是诺亚的个人资料。他试图突出自己是一个优秀的领导者，提到他在远程和办公室领导团队方面的卓越表现，涉及数据科学和机器学习，并部署了可扩展的AI解决方案。但这些都与诺亚想申请的职位无关。然而，如果仔细查看，会发现诺亚也具备一些相关技能，如熟悉Ruby、Python、JavaScript，并且有18年的软件工程师经验。
如何帮助诺亚在简历中突出这些技能，从而增加他获得面试机会的可能性呢？在我们的课程结束时，你将能够构建一个多智能体系统来实现这一目标。我们将利用四个不同的智能体：技术职位研究员、个人资料优化工程师、简历策略师和面试准备员。这四个智能体将利用从互联网搜索到简历优化的各种工具，帮助诺亚和你自己突出合适的技能以匹配职位要求。
来看一下改进后的个人资料。新的简历更加注重诺亚的技术技能，如JavaScript、Python、Ruby，并提到了UI/UX以及HTML和CSS。这些都是符合职位要求的技能，使得诺亚更有可能获得面试机会。如果你想构建这样的系统，请继续关注，因为课程结束时，你不仅可以构建这个系统，还可以构建更复杂的自动化智能体系统。
关于自动化技术的演变，过去的自动化是从点A到点B的简单编程，随着边界情况的增加，代码变得越来越复杂。而现在的智能体自动化则不同，你不需要绘制复杂的路径图，而是展示选项，这是一种全新的软件编写方式。
传统应用程序有明确的输入和输出，而在新的AI应用中，输入和输出都是模糊的。用户输入可能是字符串，但你不知道它是表示表格数据、Markdown、普通文本还是数学运算。转化过程也是模糊的，因为大语言模型（LLM）可能将其转化为列表或完整段落。因此，输出形式也是多样的，这正是其魅力所在。
举例来说，在数据收集和分析方面，传统方法通过表单捕获潜在客户数据，并通过自动化过程对其进行分类。然而，现在你可以使用多智能体系统来超越传统方法。你可以让智能体进行研究、收集数据、比较公司并进行评分，最终生成与客户对话的要点，然后将这些信息传递给销售团队，从而显著提升数据质量。
总的来说，智能体是一种能够自我决策和执行任务的软件单元，在这门课程中，我们将深入了解它们的具体运作机制。
课程原文字幕
I'm so excited to dive into this with you. We're going to learn everything that needs to know about agents and in this first lesson we're going to understand what make them tick, how they work and even look at some of the examples that we're going to do together throughout the entire course. So let's dive in real quick. So in this course, we're going to talk about so many different things, so many interesting concepts. I want to just highlight a few so that you know what you're going to see. We're going to be talking about role-playing. We're going to talking about agents. They're able to focus, to use tools to cooperate with each other. We're going to really dive into how guard rails are crucial to make agents work and how memory can make them so much better. We're going to talk about all the different ways that agents can collaborate, not only working sequentially and hierarchically, but also asynchronously.
{ 0:54 }
There's so many different options, so make sure that you stick around because we're going to build a lot of different crews throughout the different lessons. A crew is a team of AI agents working together, each with their own defined roles. Don't worry if you don't know what this means yet, That's what this course is about. We're going to start with an overview where we're going to build a simple research and right crew, but we're going to go all the way to build more complex ones, building a customer support crew, a customer outreach crew, and even planning agent system and a financial analysis agent system. And then to wrap up, we're going to build our most complex crew yet, a crew that is going to be able to tailor made your resume to any job posting out there, increasing your odds of getting an review. So make sure to stick around for that because honestly, things are only going to get more interest from here out. So what we will build in this course, Let's look at this full stack engineering job posting. So you can see there is a bunch of interesting stuff in here.
{ 1:59 }
If you look close at the job description, you're going to notice that they're basically looking for a full stack developer. They want someone that nails both front and back end. And there is a bunch of like actual criteria in there. They want you to be able to write APIs and to have experience with databases and all that. So this is what Noah's profile looks like. I'm not going to read over it, but let's highlight some of the stuff. You can see that Noah is trying to highlight how he's a good leader. So he mentioned how he has excelled in leading teams, both remote and in office. He mentioned some data science and machine learning, and he also mentions about deploying scalable AI solutions. That is nothing to do with the actual position that Noah wants to apply. But if you dig in there, you're going to notice that Noah also has some of those skills.
{ 2:53 }
You can see that mentions there that it knows Ruby, Python, JavaScript, that it has been a software engineer for 18 years. So there is definitely enough things in there that kind of could explore and that make Noah qualify for this position. So how do we make sure that we help Noah to highlight out that in his resume so that he increases his odds of getting an actual interview for the position that he does wants to apply? Well, you can use agents to do it in our case, and by the end of this course, you're going to be able to put together a multi agent system that does that. And that we do that by leveraging 4 different agents. It's going to have a tech job researcher, a personal profile for engineer, a resume strategist for engineers and engineering interviewer preparer.
{ 3:44 }
Together, those four agents will have a few tools that they're going to be able to use from searching the Internet all the way to do rag over your resume, everything to kind of like help in this case or friend Noah, but yourself as well to make sure that you're highlighting the right skills for the job. So this is what the new profile looks like. Let's compare both for a second. So in the second one, you can see that it was highlighting a lot of his leading experiences. But now in the second one, it's kind of like doubling down on the other skills that he has that better match the job that he's applying to. You can see that we mentioned JavaScript, Python, Ruby. We also mentioned UI, UX. You mentioned how he knows HTML and CSS. It's everything in there. So it's still the same profile and the same SKU set, but framing a way that now better allows them, or in this case, allows Noah to get the interview that he desires. So stick around if you want to build this, because by the end of it, you're going to be able to build not only this, but way more complex groups of the agents. They're going to be able to do a bunch of automation for you. All right, So what is a Gen. tech automation?
{ 4:57 }
If you think about what automation used to be like before it was something completely different in the past, you would say, hey, I want to go from point A to point B and you can write code to automate that. And then what happens is that as edge cases appears, you start to kind of make that a little more complex. You start to add a lot of conditions. And if analysis there where like, well, if X do C, if Z do D, and you can see how these things can become quite complex, especially as you start to adding more and more edge cases. So in the old days, whenever you're trying to do automations, what you end up with is this very complex code base with a lot of different conditions and ash cases and you just can't never cover it all. The beauty of agentic automations is that you don't need to drown the map, you can show the options. So it's fundamentally a new way to write softer.
{ 5:58 }
All right, So in regular applications you have strong inputs, so you know exactly the data that is get into your application. You understand if that's a string, if that's an integer, if that's a float, you have a very clear understanding of that. And then you also have a very clear understanding on what is the mathematical operation that you might be doing with that or any other heuristics. So you know, if you're multiplying those numbers together, if you're interpolating their string, whatever is happening in there, you have fully control and understanding of what it is. And because of that, you also understand what is the output. You know, if the output is going to be another float or another integer or another string, you can replicate it. And that's the beauty of regular engineering. But with this new AI applications, you have something else differently. You have a fuzzy input, meaning that you don't know why the user is input into your application.
{ 6:57 }
O you know that it's a string because we're talking about an LLM, but you don't know if it's a string that represents tabular data, if it's a markdown, if it's a regular text, or if it's a math operation. And then the transformations are also fuzzy because they're LLMS. You don't know if the LLM will decide to transform this into a list or if it's right, a full blown paragraph. And because of that, you don't know exactly what the output will be because it can take different forms and shapes depending on the inputs and on the transformations. But that's the beauty of it. There is a place in the world right now where these fuzzy applications make way more sense than regular existing software applications. So that's the main reason why people like ChatGPT so much. If you think about ChatGPT, that's an AI application with fuzzy inputs. You don't know what your users putting in the chat. It's also fuzzy transformations.
{ 7:59 }
You don't know what the LLM will do with that data and it's also fuzzy outputs. You don't know what will be the final output, what will be the final response for that user input. And that's why people love it, because it's something they can relate to. If you think about the word as we have today, the word is a fuzzy place. So this is the beauty about AI applications is that it's available too. That has its merits and places where you want fuzzy applications and where you want strong typed applications. And each one has its merits and its place depending on the software that you're trying to build. So let's look at some actual example here. Let's talk about data collection and analysis. And if you have been working in engineering for a while, you're probably familiar with this. Your company probably has a website that captures leads data through a form. And those leads eventually go into your sales team that are going to try to make them into customers. So this is very usual for most of the businesses out there.
{ 8:59 }
And the way that these leads become customers is by prioritizing them. And there's a bunch of ways that companies used to do that. So you have your form and you capture information about this company through through that form and then goes through an automated process of classifying these leads where you might look at some data points like, hey, is this company a big company? It has more than 10 employees. Is this company located in the US or is this company located somewhere else? And depending on the answers for those questions, you might give a different score for these companies or treat them differently on your sales process. So this is regular data analysis and data collection. But it turns out that now that you can use multi agent systems, you can go beyond that and you can go even further. So let's try something different.
{ 9:51 }
Let's think how a crew of AI agents could actually help us here and do something that's better than what we have been doing through the past few years. Let's add a crew of AI agents on the lead generation process. So what you can do now instead of having all those IFS, analysis, analyzing like specific data points for the companies, you could have this AI agents do research, go out there and do data collection. And those research can be searching Google can be searching online, can be searched on an internal database or internal data set that you might have, whatever you want, any place where you can find more information about these companies. Then you can also have these agents drawn comparisons between these companies and companies that you already have on your data set, companies that you're already talking to, companies that turned out to be great customers. And then you can also have this AI agents do some scoring so they can actually score this company and making sure that this company has a score that you can actually prioritize and decide to who you're going to be sending it to.
{ 10:58 }
And then the final point point is coming up with talking points. So not only doing the research, the comparison, the scoring, but also getting ahead and making sure that you know what are the topics that you should bring up when you go out and reach out to this customer and what that conversation should look like. And now you can take that and send that to your sales team. And you have a way better data set to work with than the one you used to have before. So this is a great example on how regular automation can actually become way better by using a Gen. tech automation. All right, so we're talking a lot about agents so far. We're talking about a lot of different types of automations. But in the end of the day, what are agents? So let's talk about that for a second and step back.
有关Notion安装或者使用上的问题，欢迎您在底部评论区留言，一起交流~
- 作者:luxinfeng
- 链接:https://www.luxinfeng.top/article/%E8%AF%BE%E7%A8%8B%E6%80%BB%E8%A7%88
- 声明:本文采用 CC BY-NC-SA 4.0 许可协议，转载请注明出处。